CEAP Reimbursements descriptive analysis

This notebook presents a descriptive analysis of CEAP expenses addressing the following questions:

  1. Monthly average per congressperson (grouped per state, as values varies across states)
  2. Monthly ratio of the total allowed and the total spent per congressperson (grouped per state, as values varies across states)
  3. Monthly average per subquota (grouped per state, as values varies across states)

Questions are not explicitly answered. Charts and tables are provided for free interpretation, some of them with brief commentaries from the author. Obs.: original analysis was made considering data from 2009 to 2017. One might want to filter by terms (e.g. 2010-2014) to make more realistic comparisons (spenditures by state, congressperson, subquota).

In [1]:
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
import seaborn as sns
In [2]:
reimbursements = pd.read_csv('../data/2017-07-04-reimbursements.xz',
                   parse_dates=[16],
                   low_memory=False,
                   dtype={'document_id': np.str,
                          'congressperson_id': np.str,
                          'congressperson_document': np.str,
                          'term_id': np.str,
                          'cnpj_cpf': np.str,
                          'reimbursement_number': np.str,})
reimbursements.shape
Out[2]:
(1619213, 31)
In [3]:
reimbursements.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1619213 entries, 0 to 1619212
Data columns (total 31 columns):
year                          1619213 non-null int64
applicant_id                  1619213 non-null int64
document_id                   1619213 non-null object
reimbursement_value_total     1619213 non-null float64
total_net_value               1619213 non-null float64
reimbursement_numbers         1619213 non-null object
congressperson_name           1619213 non-null object
congressperson_id             1616157 non-null object
congressperson_document       1616157 non-null object
term                          1619213 non-null float64
state                         1616157 non-null object
party                         1616157 non-null object
term_id                       1616157 non-null object
subquota_number               1619213 non-null int64
subquota_description          1619213 non-null object
subquota_group_id             1619213 non-null int64
subquota_group_description    1619213 non-null object
supplier                      1619213 non-null object
cnpj_cpf                      1618012 non-null object
document_number               1619204 non-null object
document_type                 1619213 non-null int64
issue_date                    1619213 non-null object
document_value                1619213 non-null float64
remark_value                  1619213 non-null float64
net_values                    1619213 non-null float64
month                         1619213 non-null int64
installment                   1619213 non-null int64
passenger                     5765 non-null object
leg_of_the_trip               17965 non-null object
batch_number                  1619213 non-null int64
reimbursement_values          1619213 non-null float64
dtypes: float64(7), int64(8), object(16)
memory usage: 383.0+ MB

congressperson_name has 1619213 entries, but congressperson_document has fewer entries: 1616157. This is because some reimbursements was made to the parties. As you can see below.

In [4]:
reimbursements[reimbursements["congressperson_document"].isnull()]["congressperson_name"].unique()
Out[4]:
array(['LIDERANÇA DO PT', 'LIDERANÇA DO PSDB', 'PDT', 'SDD', 'PROS',
       'LID.GOV-CD', 'PHS', 'PTN', 'PSOL'], dtype=object)
In [5]:
reimbursements["year_month"] = reimbursements["year"].astype(str) + "/" + reimbursements["month"].astype(str)
In [6]:
reimbursements.sort_values(by=["year_month","state"], ascending=[False,True], inplace=True)
In [7]:
states = reimbursements[reimbursements["state"].notna()]["state"].unique()
In [8]:
year_months = reimbursements[reimbursements["year_month"].notna()]["year_month"].unique()

Filters depending on the scope of analysis

Here, filters by state, party, years, etc. can be applied.

Obs.: chart commentaries provided might not remain valid depending on filters chosen.

In [9]:
# Filters only most recent years (from 2015)
# reimbursements = reimbursements[(reimbursements['year'] == 2015) | (reimbursements['year'] == 2016) | (reimbursements['year'] == 2017)]

Defining ploting functions

In [10]:
#This function plots values grouped by states sharing the same x axis.
def plot_data_same_axis(data, size, aspect, x_column, y_column, x_label, y_label):
    g = sns.FacetGrid(row="state", data=data, palette="YlOrRd", sharey=False, size=size, aspect=aspect)
    g.map(sns.barplot, x_column, y_column, palette="YlOrRd")
    g.set_axis_labels(x_label, y_label)
    for ax in g.axes.ravel():
        for p in ax.patches:
            ax.text(p.get_width(), p.get_y()+0.5, '%1.2f' % p.get_width(), fontsize=10, color='black', ha='left', va='center')
    plt.show()
In [11]:
#This function plots independents charts with values grouped by states. The charts do not share the axis or sizes.
def plot_data_diff_axis(data, x_column, y_column, x_label, y_label):
    for state in states:
        selection = data[data.state == state].sort_values(['total_net_value'])
        if len(selection) > 35:
            plt.figure(figsize=(10,25))
        ax = sns.barplot(y=y_column, x=x_column, data=selection, palette="YlOrRd")
        for p in ax.patches:
            ax.text(p.get_width(), p.get_y()+0.5, '%1.2f' % p.get_width(), fontsize=10, color='black', ha='left', va='center')
        ax.set_title(state)
        ax.set(ylabel=y_label, xlabel=x_label)
        plt.show()

1. Monthly average per congressperson

Monthly average per congressperson (grouped per state, as values varies across states)

In [12]:
states
Out[12]:
array(['AC', 'AL', 'AM', 'AP', 'BA', 'CE', 'DF', 'ES', 'GO', 'MA', 'MG',
       'MS', 'MT', 'PA', 'PB', 'PE', 'PI', 'PR', 'RJ', 'RN', 'RO', 'RR',
       'RS', 'SC', 'SE', 'SP', 'TO'], dtype=object)
In [13]:
monthly_avg_by_month = reimbursements.groupby(['state', 'congressperson_name', 'year_month']).agg({'total_net_value': 'mean'})
In [14]:
monthly_avg = monthly_avg_by_month.groupby(['state', 'congressperson_name']).agg({'total_net_value': 'mean'})

Ploting data

In [15]:
data = pd.DataFrame(monthly_avg.to_records()).sort_values(['state','total_net_value'])
In [16]:
plot_data_diff_axis(data,'total_net_value', 'congressperson_name', 'Monthly Avg (R$)', 'Congress Person')

/opt/conda/lib/python3.6/site-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use.
  stat_data = remove_na(group_data)
In [17]:
#plot_data_same_axis(data, 30, 1, 'total_net_value', 'congressperson_name', 'Monthly Avg (R$)', 'Congress Person')

2. Monthly ratio of the total allowed and the total spent

Monthly ratio of the total allowed and the total spent per congressperson (grouped per state, as values varies across states)

Computing estimated CEAP values for years in between 2009 and 2017 using CAGR

thanks to luipillmann for the notebook 2017-05-01-luipillmann-intro-to-reimbursements.ipynb!

In [18]:
# Source 2017-05-01-luipillmann-intro-to-reimbursements.ipynb. Thanks!

# Source: http://www2.camara.leg.br/a-camara/estruturaadm/diretorias/dirgeral/estrutura-1/deapa/portal-da-posse/ceap-1
ceap_2009 = [40711.32, 37318.73, 39734.17, 39554.50, 35540.51, 38705.50, 27977.66, 34080.83, 32317.69, 38429.49, 32856.38, 36949.65, 35924.24, 38499.17, 38319.91, 37992.68, 37344.18, 35412.67, 32550.32, 38963.25, 39828.33, 41612.80, 37256.00, 36337.92, 36578.43, 33730.95, 35993.76]

# Source: http://www2.camara.leg.br/comunicacao/assessoria-de-imprensa/cota-parlamentar
ceap_2017 = [44632.46, 40944.10, 43570.12, 43374.78, 39010.85, 42451.77, 30788.66, 37423.91, 35507.06, 42151.69, 36092.71, 40542.84, 39428.03, 42227.45, 42032.56, 41676.80, 40971.77, 38871.86, 35759.97, 42731.99, 43672.49, 45612.53, 40875.90, 39877.78, 40139.26, 37043.53, 39503.61]

ceap_limit_states = pd.DataFrame({
    '2009': ceap_2009,
    '2017': ceap_2017
}, index=states)

# Computes estimated CEAP values for years in between 2009 and 2017 using CAGR
ceap_values = pd.DataFrame(columns=['year', 'state', 'quota'])

ini_year = 2009
for i in range(2018-ini_year):
    year = ini_year + i
    state_index = 0
    for state in states:
        quota = -1
        if year == 2009:
            quota = ceap_limit_states['2009'][state_index]
        elif year == 2017:
            quota = ceap_limit_states['2017'][state_index]
        else:
            # Calculates CAGR according to data available (CEAP@2009 and CEAP@2017), using the CEAP per state
            cagr = ((ceap_limit_states['2017'][state_index] / ceap_limit_states['2009'][state_index])**(1./(2017-2009)) - 1)
            quota = ceap_values[(ceap_values['state'] == state) & (ceap_values['year'] == year-1)]['quota'].values[0] * (1 + cagr)
        ceap_values = ceap_values.append(pd.DataFrame([[year, state, quota]], columns=['year', 'state', 'quota']))
        state_index += 1
        
ceap_values.head(2)
Out[18]:
year state quota
0 2009 AC 40711.32
0 2009 AL 37318.73
In [19]:
monthly_sum = reimbursements.groupby(['state', 'congressperson_name', 'year_month']).agg({'total_net_value': 'sum'})
In [20]:
def calc_ratio(state, name, year_month, total_net_value):
    year = int(year_month.split('/')[0])
    ratio = (total_net_value / ceap_values[(ceap_values['state'] == state) & (ceap_values['year'] == year)]['quota'].values[0])
    #print(state+' '+year_month+' '+str(ratio))
    return ratio
In [21]:
monthly_sum_with_ratio = pd.DataFrame(monthly_sum.to_records())
monthly_sum_with_ratio['ratio'] = monthly_sum_with_ratio.apply(lambda x: calc_ratio(*x), axis=1)
In [22]:
monthly_avg = monthly_sum_with_ratio.groupby(['state', 'congressperson_name']).agg({'ratio': 'mean'})

Ploting the data

In [23]:
data = pd.DataFrame(monthly_avg.to_records()).sort_values(['state','ratio'])
In [24]:
#plot_data_diff_axis(data,'ratio', 'congressperson_name', 'Monthly Ratio Avg (R$)', 'Congress Person')
In [25]:
plot_data_same_axis(data, 30, 1, 'ratio', 'congressperson_name', 'Monthly Ratio Avg (R$)', 'Congress Person')

/opt/conda/lib/python3.6/site-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use.
  stat_data = remove_na(group_data)

3. Monthly average per subquota

Monthly average per subquota (grouped per state, as values varies across states)

In [26]:
monthly_avg_by_month = reimbursements.groupby(['state', 'subquota_description', 'year_month']).agg({'total_net_value': 'mean'})
In [27]:
monthly_avg = monthly_avg_by_month.groupby(['state', 'subquota_description']).agg({'total_net_value': 'mean'})

Ploting the data

In [28]:
data = pd.DataFrame(monthly_avg.to_records()).sort_values(['state','total_net_value'])
In [29]:
plot_data_same_axis(data, 4, 2, 'total_net_value', 'subquota_description', 'Monthly Avg (R$)', 'Subquota Description')

/opt/conda/lib/python3.6/site-packages/seaborn/categorical.py:1428: FutureWarning: remove_na is deprecated and is a private function. Do not use.
  stat_data = remove_na(group_data)